Neural Networks Designing Research Based On Cubic Spline Interpolant In Power Exponent Weight Functions

Based on the feature of Artificial Neural Network availablely and the theories of interpolant function in power exponent ,we put forward the new concept of weighting function of Neural Network which is about of the input sample data. And we brought forward a new Artificial Neural Network successfully, i.e. cubic spline interpolant in power exponent weight functions neural networks. In this paper the weighting function of trained network is cubic spline interpolant in power exponent function which is different from the weight constant of classical networks. The new Neural Network is that the input layer connects with nerve cells directly and the trained weight is weighting function of the input sample data. Its input and output form is: the input node number of network equals with the dimensions of input vector m, the output node number of network equals with the dimensions of output vector n, and it outputs the trained result of j(j=1,2,…,n) nerve cell directly.The new networks algorithm overcomes deficiencies of classical feed forward networks i.e. BP algorithm and RBF networks of ANN of which the weight did not reflect the information of input samples, and as the number of samples increased, the generalization of network enhances. Besides, the structure of new network is simple, so the computation is only same with solving linear equations. Eventually, the computer imitating example show that the neural network has a better nonlinear reflection ability, and can describe the complex relationship between the independent variable and the dependent variable with better precision, and has well feasibility.Also, this paper gives the steps of cubic spline interpolant in power exponent weight functions neural networks' algorithm and detailedly recites the procedure of data proceeding.